Yup. It's a good testing and development environment. What if the use case is 50-50 windows Mac?
Say I need to run MATLAB on windows, and Photoshop on Mac. Parallels? Don't think it'll work well unless they're trivial simulations. I've seen complex excel worksheets accessing datasets fail in parallels. And that's with a 16gb machine. I've also seen windows updates fail.
Parallels is a good solution for many use cases. But it's not a miracle pill.
Nothing is 100%. I'm sure eventually you will find certain software that have requirements that doesn't fit into a machine which you have previously spec'ed.
I don't know what spec'ed of the machine that you are on and what version of the Parallels Desktop that you are using that resulted in the failed scenarios that you have provided. I have in totality of 3 different OSes on top of the Mac OS Maverick, 1x Windows XP, 1x Ubuntu Server, 1x Mac OS X El Capitan all at once and they are pretty functional albeit slower.
My laptop is an old Macbook Pro 17"(2009) on 8GB, 3.06Ghz Core 2 Dual on a Samsung Evo 840 512GB system.
I don't think your Windows updates must be attributed to Parallels Desktop. Can you be absolutely sure it is not because the state at which the Windows guest that contributed to the failure, and not because it is running on Parallels Desktop ? Not to isolate that a hypervisor can contribute to such an issue, but so many years I have been using Parallels Desktop(since version 3, to now version 11), from Windows XP till to Windows 10, even Windows Servers, Linux on Ubuntu, Redhat, CoreOS and more, even Mac OS. None of them have shown any updates failure except performance slowness at most.
As for your complex excel worksheet, what kind of failures did you encountered. What is the size at which they failed ? Can you at least be specific so that we can have a better assessment ?
If one requires the most performance from a machine, any virtualise scenarios doesn't fit in, but the question here is across the board, how much of the use cases that claimed to failed are indeed purely due to virtualisation and not because of something else ? Compared to the benefits you can reap from a virtualise environment, would such edge cases be acceptable ?
BTW I would also like the audiences to take a better look into situations when one experience slowness in a machine. Take a look at benchmarks done on the latest VM hypervisors
For eg:
https://www.tekrevue.com/2015-vm-benchmarks-parallels-11-vs-fusion-8/9/
This is a benchmark on Handbrake across bootcamp, a native installation of Windows and with all other VM hypervisors. Besides Virtualbox which is not the most optimised solution, you find both VMWare fusion and Parallels Desktop, trailing behind bootcamp just marginally. The benchmark CPU bounded. That is to say when it comes to raw computing power, today's VM hypervisors are very optimised and the overhead is very low. Most of the time, the slowness are in the graphical department which trails with approximately 40% overhead for Parallels Desktop and marginal for VMWare Fusion. If 3D graphical performance in the guest is what you need, you can choose VMWare Fusion instead.
Source:
https://www.tekrevue.com/2015-vm-benchmarks-parallels-11-vs-fusion-8/5/
https://www.tekrevue.com/2015-vm-benchmarks-parallels-11-vs-fusion-8/5/
When it comes to office suite, it is roughly a 15% overhead. Productivity tools normally don't require much cpu resources, albeit I notice they are more and more graphically hungry especially with the most recent version of M$ Office.
With the earlier example you have provided, photoshop in mac and matlab in windows. The former is largely graphical intensive (less the filters which are mostly CPU intensive), running in Mac will be fine. Your MATLAB should be cpu intensive instead of graphical, so with marginal fall back, it shouldn't be that bad running in a VM guest. That is to say, if you reach system bottleneck with these 2 combination, it will most likely make little differences even if both are running on native Windows platform.